Potentially missing data are considerably more frequent than definitely missing data: a methodological survey of 638 randomized controlled trials
Journal Articles
Overview
Research
Identity
Additional Document Info
View All
Overview
abstract
BACKGROUND AND OBJECTIVE: Missing data for the outcomes of participants in randomized controlled trials (RCTs) are a key element of risk of bias assessment. However, it is not always clear from RCT reports whether some categories of participants were followed-up or not (i.e., do or do not have missing data) nor how the RCT authors dealt with missing data in their analyses. Our objectives were to describe how RCT authors (1) report on different categories of participants that might have missing data, (2) handle these categories in the analysis, and (3) judge the risk of bias associated with missing data. METHODS: We surveyed all RCT reports included in 100 clinical intervention systematic reviews (SRs), half of which were Cochrane SRs. Eligible SRs reported a group-level meta-analysis of a patient-important dichotomous efficacy outcome, with a statistically significant effect estimate. Eleven reviewers, working in pairs, independently extracted data from the primary RCT reports included in the SRs. We predefined 19 categories of participants that might have missing data. Then, we classified these participants as follows: "explicitly followed-up," "explicitly not followed-up" (i.e., definitely missing data), or "unclear follow-up status" (i.e., potentially missing data). RESULTS: Of 638 eligible RCTs, 400 (63%) reported on at least one of the predefined categories of participants that might have missing data. The median percentage of participants who were explicitly not followed-up was 5.8% (interquartile range 2.2-14.8%); it was 9.7% (4.1-14.9%) for participants with unclear follow up status; and 11.7% (interquartile range 5.6-23.7%) for participants who were explicitly not followed-up and with unclear follow-up status. When authors explicitly reported not following-up participants, they most often conducted complete case analysis (54%). Most RCTs neither reported on missing data separately for different outcomes (99%) nor reported using a method for judging risk of bias associated with missing data (95%). CONCLUSION: "Potentially missing data" are considerably more frequent than "definitely missing data." Adequate reporting of missing data will require development of explicit standards on which editors insist and to which RCT authors adhere.
